Electrodynamic sensors and neural networks for electrical charge tomography

• Main Author: Bidin, Abdul Rahman
• Published: Sheffield Hallam University 1993
• Subjects: 621.381045; Optoelectronics
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.334572

This research into the feasibility of imaging particulate processes using electrical charge tomography investigates four techniques: the multi-sensing of electrical charge in a cross-section, a neural network based classifier for flow regime identification, cross correlation based velocity determination and spectral analysis of electrodynamic signals. A single charged-particle model is developed to simulate the induction effect on a sensor by a charge. The spatial representation of the voltage induced onto sixteen sensors, placed on the boundary of a circular pipe, gives a flow distribution profile over the cross-section. A two charged-particle model is developed to simulate the electrodynamic effect of two particles on a tomographic sensor configuration. As in the single particle model, a spatial representation of the voltages induced onto the sensors is presented. This voltage profile is due to the combined effects of position and charge of the two particles. A multi-particle model is developed to predict the voltage profile of several flow regimes: full, annular, core, half and stratified. The model is extended to provide the loading and concentration of a given flow. A measurement system is constructed consisting of sixteen sensors equally spaced around the boundary of a circular 100mm pipe. Measurements on a bead drop system are designed to verify the single particle model. A sand flow system, consisting mainly of 300 micron sized particles, is used for measurements of the induced voltages due to different flow regimes. The latter are created artificially by using baffles of different shapes that obstruct the sand flow. The voltage profile from the sixteen sensors gives spatial information about the flow regime. These voltage profiles are normalised into patterns that are presented to a Kohonen neural network for classification. Two regime classification between well differentiated regimes gives an accuracy of identification of 95%. This is expanded to provide classification of three regimes with more variability in the input patterns giving success rates between 50% to 70%. A power spectral density analysis of the measured electrodynamic signals gives observable features for particle characterisation during flow. In full flow, with no baffles obstructing the sand flow, a consistently high frequency spectra of 550Hz is observed. At flow rates above 0.540 kgs-1, the frequency spectra shifts to a lower range of 200Hz. In obstructed flow, such as in stratified regime, an inhomogeneous phase is inferred from the drop in frequency of the power spectra at relatively low flowrates (0.36kgs-1). These results suggest a relationship between the observed spectra and the phenomenon of clustering of particles at higher concentrations. The potential of electrodynamic spectroscopy for particle characterisation in terms of size distribution is discussed. Knowledge of flow regime voltage profile, regime identification and concentration provided a basis for an empirically based image reconstruction algorithm. Finally the achievements of the thesis are discussed and suggestions made for further work.